3 ms·

Much as I am in support of open source, this isn't a magic cure. I just got confirmation that POWER is susceptible to Meltdown/Spectre so there's that...

rbanffy 9y ago
But, at least, if the architecture (not just the ISA) are open, the safety can be verified and, if we are lucky, even proven. BTW, it seems SPARC (also open) is not vulnerable.
- ThrowawayR2 9y ago
  > ...if the architecture (not just the ISA) are open, the safety can be verified... That's just the old "many eyes make all bugs shallow" fallacy, which has long been discredited. Given that architects at nearly all of the processor makers on the planet missed Spectre and the article itself notes that only "chip weenies" (their phrasing, not mine) could've figured it out, there's no real basis for such claims. > BTW, it seems SPARC (also open) is not vulnerable. OpenSPARC happened to be open and happened to be not vulnerable, therefore the former caused the latter? Unless there's some persuasive evidence otherwise, that's clearly faulty reasoning.
  - rbanffy 9y ago
    The facts behind Meltdown and Spectre are not new. We all knew not all side effects of speculative execution were reversed and some could even be impossible to (even if you clear a cache, if a read went to the bus, the CPU already disclosed to an external observer an execution path that was later discarded). What happened is that someone insanely clever figured out a way to exploit those side effects we all thought were innocuous. In the future, when someone looks at an open CPU design and spots an out of order core that leaks side effects of instructions that were on branches not taken, they'll hopefully remember this mess and proceed with care. Processor erratas are always a worthy, even if sometimes terrifying, read. And even if we consider the eyeballs are not optimally distributed, the more eyeballs we have, the shallower the bugs will be. Like I mentioned, we all knew (and I'm not even actively working in chip design - I'm just someone who reads research papers when I can) some speculative execution side-effects remain after the CPU rewrites the past and discards the primary effects.

rbanffy 9y ago

But, at least, if the architecture (not just the ISA) are open, the safety can be verified and, if we are lucky, even proven. BTW, it seems SPARC (also open) is not vulnerable.

ThrowawayR2 9y ago
> ...if the architecture (not just the ISA) are open, the safety can be verified... That's just the old "many eyes make all bugs shallow" fallacy, which has long been discredited. Given that architects at nearly all of the processor makers on the planet missed Spectre and the article itself notes that only "chip weenies" (their phrasing, not mine) could've figured it out, there's no real basis for such claims. > BTW, it seems SPARC (also open) is not vulnerable. OpenSPARC happened to be open and happened to be not vulnerable, therefore the former caused the latter? Unless there's some persuasive evidence otherwise, that's clearly faulty reasoning.
- rbanffy 9y ago
  The facts behind Meltdown and Spectre are not new. We all knew not all side effects of speculative execution were reversed and some could even be impossible to (even if you clear a cache, if a read went to the bus, the CPU already disclosed to an external observer an execution path that was later discarded). What happened is that someone insanely clever figured out a way to exploit those side effects we all thought were innocuous. In the future, when someone looks at an open CPU design and spots an out of order core that leaks side effects of instructions that were on branches not taken, they'll hopefully remember this mess and proceed with care. Processor erratas are always a worthy, even if sometimes terrifying, read. And even if we consider the eyeballs are not optimally distributed, the more eyeballs we have, the shallower the bugs will be. Like I mentioned, we all knew (and I'm not even actively working in chip design - I'm just someone who reads research papers when I can) some speculative execution side-effects remain after the CPU rewrites the past and discards the primary effects.